Social Network Analysis
Introduction
What Question Does It Ask?
Social network analysis studies the graph formed by treating people as nodes and relationships as edges. The question is not the traits of individuals but the structure of their ties: who is central, which groups exist, and how a rumor or a disease spreads through the network.
At a glance
graph theory and statistics
nodes and links
centrality, communities, diffusion
modularity, interpretability
medium: graphs, matrices
network data
Introduction
A Short History
Euler solves the Königsberg bridges problem; graph theory is born.
Moreno draws relationships in diagrams he calls sociograms.
Milgram's letter experiment popularizes the idea of "six degrees".
Granovetter shows "the strength of weak ties": job leads come from distant acquaintances.
Watts and Strogatz publish the small-world network model.
The Louvain method makes fast community detection possible on networks with millions of nodes.
The Course
Core Concepts
Directed or undirected, weighted or unweighted.
A few highly connected nodes, many with few links.
Degree, betweenness, closeness and eigenvector measures.
How many of my friends know each other.
High clustering but short average paths.
Nodes densely linked inside, sparsely to the outside.
How good a partition into communities is.
Spread of information or disease on a network (SIR).
The Course
A Typical Workflow
The Course
What Does the Course Cover?
- Graph basics: nodes, edges, adjacency matrix
- Collecting network data: surveys, traces, ego nets
- Degree distribution, path length and diameter
- Centrality: degree, betweenness, closeness
- Eigenvector centrality and PageRank
- Clustering, triadic closure and weak ties
- Homophily and assortativity
- Random graphs and the small-world model
- Scale-free networks and preferential attachment
- Community detection: modularity and Louvain
- Spreading: the SIR model and information cascades
- Link prediction and network visualization
Skills you will gain
A typical semester; the order and coverage vary between universities.
Try It
Who Should Start the Campaign?
A club will hand its announcement to one person in the friendship network of its 18 members. Click a person to see their centrality, find the communities, then watch the news spread from that person step by step.
Context
Prerequisites
Courses worth knowing before this one, and why they are needed.
| Course | Why it is needed |
|---|---|
| Linear Algebra | adjacency matrix, spectral methods |
| Data Structures | adjacency lists and matrices: storing a graph in memory |
| Algorithms | BFS, DFS and shortest paths: the basis of centrality computations |
| Discrete Mathematics | graph theory, sets and counting |
Context
Are You Ready for This Course?
Ask yourself; for every "no", the right-hand column shows where to start.
| Ask yourself | If not, start with |
|---|---|
| Can you find the shortest path between two nodes of a graph with BFS? | Graph search from an algorithms course |
| Can you explain what an eigenvector of a matrix means? | Linear Algebra |
| Can you compute the mean and variance of a distribution? | Probability and Statistics |
| Can you read an edge-list file into a Python dictionary? | An intro programming course, then a short NetworkX tutorial |
Context
Neighbouring Courses
Overlap
Overlapping Courses
| Relation | Course | Shared ground |
|---|---|---|
| builds on | This course builds on Linear Algebra | adjacency matrix, spectral methods |
| shares methods | Shares methods with Data Mining | community detection |
| shares methods | Shares methods with Information Retrieval and Recommender Systems | PageRank, link analysis |
Differences
Where It Differs (1/2)
| Criterion | Social Network Analysis | Data Mining | IR & Recommenders |
|---|---|---|---|
| Core question | How are networks of people and their relationships analyzed? | How are hidden, useful patterns found in large piles of data? | How are the most relevant documents or products picked from millions of candidates? |
| Approach | graph theory and statistics | discovery-driven analysis | matching and ranking |
| Input | nodes and links | large tables or transaction data | query, documents, user history |
| Output | centrality, communities, diffusion | rules, clusters, outliers | ranked result list |
| Success measure | modularity, interpretability | interestingness, support, confidence | precision, recall, NDCG |
| Math load | medium: graphs, matrices | medium | medium: vector spaces |
| Data needs | network data | large | lots of interaction data |
- Data Mining: Data mining looks for patterns in tabular records; this course studies the structure of the links between records.
- IR & Recommenders: Information retrieval uses PageRank as a ranking signal; this course asks what centrality says about the network.
Differences
Where It Differs (2/2)
| Criterion | Social Network Analysis | Linear Algebra | Multi-Agent Systems |
|---|---|---|---|
| Core question | How are networks of people and their relationships analyzed? | How do vectors and matrices represent data and transformations? | How do several autonomous agents decide together or against each other? |
| Approach | graph theory and statistics | vector spaces and transformations | interacting autonomous agents |
| Input | nodes and links | vectors, matrices | agents, goals, protocols |
| Output | centrality, communities, diffusion | solution, decomposition, transformation | joint behavior, equilibrium |
| Success measure | modularity, interpretability | accuracy, numerical stability | social welfare, stability |
| Math load | medium: graphs, matrices | the theory itself | medium: game theory |
| Data needs | network data | none | little; simulation |
- Linear Algebra: Linear algebra computes the eigenvalues of the adjacency matrix; this course reads them as communities and importance.
- Multi-Agent Systems: Multi-agent systems model how agents make decisions; this course analyzes the observed network of ties.
Practice
Applications
Influencers and echo chambers.
Disease spread on contact networks.
Rings of linked fake accounts.
Citation and co-authorship analysis.
How knowledge flows in a company.
Link prediction from mutual friends.
Practice
Risks and Ethics
A friendship network can reveal political views or sexual orientation that a person never disclosed.
Even with names removed, the pattern of links can identify a person, as structural re-identification attacks showed.
In the Cambridge Analytica case, a personality quiz app also collected data on its users' friends without their consent.
Ties forgotten in a survey or invisible on a platform can badly distort centrality rankings and communities.
In depth: Explainable AI and AI Ethics.
Practice
Tools and Datasets
Tools and libraries
Well-known datasets
Practice
Common Misconceptions
| Common belief | In fact |
|---|---|
| The person with the most friends is the most influential. | Degree is one measure; a bridge joining two groups (betweenness) can matter more. |
| Looking at a network drawing is enough to understand it. | Large networks draw as a "hairball"; numerical measures and tests are needed. |
| Community detection has one right answer. | Different methods give different partitions; results are read with domain knowledge. |
| Connected people have influenced each other. | Similar people choose each other anyway (homophily); influence is hard to separate from selection. |
Next Steps
After This Course
Courses that build on or extend this one
Related careers
Next Steps
Further Reading
- Wasserman and Faust, Social Network Analysis: Methods and Applications, Cambridge University Press, 1994.
- Easley and Kleinberg, Networks, Crowds, and Markets, Cambridge University Press, 2010.
- Newman, Networks, 2nd ed., Oxford University Press, 2018.
- Barabási, Network Science, Cambridge University Press, 2016.